Enable LTO

examples/arduino_nano/color/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:arduino-nano</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/arduino_nano/color</option>
   </options>
   <modules>

examples/black_pill_f411/usbfatfs/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:black-pill-f411</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/black_pill_f411/usb_fatfs</option>
     <option name="modm:tinyusb:config">device.msc, device.cdc</option>
   </options>

examples/feather_m4/usbserial/project.xml

@@ -2,6 +2,7 @@
   <extends>modm:feather-m4</extends>
   <options>
     <option name="modm:tinyusb:config">device.cdc</option>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/feather_m4/usbserial</option>
   </options>
   <modules>

examples/nrf51422-dk/uart_logging/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:nrf51422-dk</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/nrf51422-dk/uart_logging</option>
   </options>
   <modules>

examples/nucleo_f031k6/sk6812/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:nucleo-f031k6</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/nucleo_f031k6/sk6812</option>
   </options>
   <modules>

examples/nucleo_f303k8/rtos/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:nucleo-f303k8</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/nucleo_f303k8/rtos</option>
   </options>
   <modules>

examples/nucleo_f767zi/ethernet/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:nucleo-f767zi</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/nucleo_f767zi/ethernet</option>
   </options>
   <modules>

examples/nucleo_f767zi/freertos_blink/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:nucleo-f767zi</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/nucleo_f767zi/freertos_blink</option>
   </options>
   <modules>

examples/nucleo_g071rb/amnb/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:nucleo-g071rb</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/nucleo_g071rb/amnb</option>
     <option name="modm:communication:amnb:with_heap">no</option>
   </options>

examples/nucleo_g071rb/apa102/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:nucleo-g071rb</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/nucleo_g071rb/apa102</option>
   </options>
   <modules>

examples/nucleo_l452re/lvgl/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:nucleo-l452re</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/nucleo_l452re/lvgl</option>
     <option name="modm:platform:cortex-m:main_stack_size">5Ki</option>
   </options>

examples/nucleo_l552ze-q/freertos_blink/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:nucleo-l552ze-q</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/nucleo_l552ze-q/freertos_blink</option>
   </options>
   <modules>

examples/samd/usbserial/project.xml

@@ -2,6 +2,7 @@
    <extends>modm:feather-m0</extends>
    <!-- <extends>modm:samd21-mini</extends> -->
    <options>
+     <option name="modm:build:lto">yes</option>
      <option name="modm:build:build.path">../../../build/samd/usbserial</option>
     <option name="modm:tinyusb:config">device.cdc</option>
    </options>

examples/weact_u585ci/usb/project.xml

@@ -1,6 +1,7 @@
 <library>
   <extends>modm:weact-u585ci</extends>
   <options>
+    <option name="modm:build:lto">yes</option>
     <option name="modm:build:build.path">../../../build/weact_u585ci/usb</option>
     <option name="modm:tinyusb:config">device.dfu_rt,device.cdc,device.msc</option>
   </options>

ext/aws/modm_port.cpp.in

@@ -26,13 +26,13 @@ void vApplicationStackOverflowHook(TaskHandle_t /*pxTask*/, char *pcTaskName)
 // Make the Newlib heap thread-safe with FreeRTOS
 
 extern "C" void __malloc_lock(struct _reent *);
-void __malloc_lock(struct _reent *)
+void modm_used __malloc_lock(struct _reent *)
 {
 	vTaskSuspendAll();
 }
 
 extern "C" void __malloc_unlock(struct _reent *);
-void __malloc_unlock(struct _reent *)
+void modm_used __malloc_unlock(struct _reent *)
 {
 	xTaskResumeAll();
 }

src/modm/architecture/utils.hpp

@@ -146,7 +146,7 @@
 	#define modm_deprecated(msg)	__attribute__((deprecated(msg)))
 	#define modm_likely(x)			__builtin_expect(!!(x), 1)
 	#define modm_unlikely(x)		__builtin_expect(!!(x), 0)
-	#define modm_section(s)			__attribute__((section(s)))
+	#define modm_section(s)			__attribute__((section(s), used))
 	#define modm_fallthrough		__attribute__((fallthrough))
 	#define modm_noreturn			__attribute__((noreturn))
 	#define modm_warn_unused_result	__attribute__((warn_unused_result))
@@ -159,7 +159,7 @@
 	 	//	- __attribute__ ((weak, alias ("__aliasedFunction"))) seems to work on Windows
 	#	define modm_weak
 	#else
- 	#	define modm_weak			__attribute__((weak))
+ 	#	define modm_weak			__attribute__((weak, used))
 	#endif
 
 	#if defined(MODM_OS_HOSTED) || defined(MODM_CPU_AVR)

src/modm/platform/heap/cortex/heap_block.cpp

@@ -55,7 +55,7 @@ void __modm_initialize_memory(void)
 extern void __malloc_lock(struct _reent *);
 extern void __malloc_unlock(struct _reent *);
 
-void* __wrap__malloc_r(struct _reent *r, size_t size)
+void* modm_used __wrap__malloc_r(struct _reent *r, size_t size)
 {
 	__malloc_lock(r);
 	void *ptr = allocator.allocate(size);
@@ -65,22 +65,22 @@ void* __wrap__malloc_r(struct _reent *r, size_t size)
 	return ptr;
 }
 
-void* __wrap__calloc_r(struct _reent *r, size_t size)
+void* modm_used __wrap__calloc_r(struct _reent *r, size_t size)
 {
 	void *ptr = __wrap__malloc_r(r, size);
 	if (ptr) memset(ptr, 0, size);
 	return ptr;
 }
 
-void* __wrap__realloc_r(struct _reent *r, void *p, size_t size)
+void* modm_used __wrap__realloc_r(struct _reent *r, void *p, size_t size)
 {
 	if (!p) return __wrap__malloc_r(r, size);
 	// NOT IMPLEMENTED!
 	modm_assert(0, "realloc", "Realloc is not implemented for Block heap!", size);
 	return NULL;
 }
 
-void __wrap__free_r(struct _reent *r, void *p)
+void modm_used __wrap__free_r(struct _reent *r, void *p)
 {
 	__malloc_lock(r);
 	allocator.free(p);

src/modm/platform/heap/cortex/heap_newlib.cpp

@@ -40,7 +40,7 @@ void __modm_initialize_memory(void)
  *  Note:  This implementation is not thread safe (despite taking a
  *         _reent structure as a parameter).
  */
-void *
+void * modm_used
 _sbrk_r(struct _reent *,  ptrdiff_t size)
 {
 	const uint8_t *const heap = heap_top;

src/modm/platform/heap/cortex/heap_tlsf.cpp

@@ -126,21 +126,21 @@ void * malloc_traits(size_t size, uint32_t traits)
 	return NULL;
 }
 
-void *__wrap__malloc_r(struct _reent *, size_t size)
+void *modm_used __wrap__malloc_r(struct _reent *, size_t size)
 {
 	// default is accessible by S-Bus and DMA-able
 	return malloc_traits(size, uint32_t(modm::MemoryTrait::AccessSBus) |
 							   uint32_t(modm::MemoryTrait::AccessDMA));
 }
 
-void *__wrap__calloc_r(struct _reent *r, size_t size)
+void *modm_used __wrap__calloc_r(struct _reent *r, size_t size)
 {
 	void *ptr = __wrap__malloc_r(r, size);
 	if (ptr) memset(ptr, 0, size);
 	return ptr;
 }
 
-void *__wrap__realloc_r(struct _reent *r, void *p, size_t size)
+void *modm_used __wrap__realloc_r(struct _reent *r, void *p, size_t size)
 {
 	if (!p) return __wrap__malloc_r(r, size);
 
@@ -156,7 +156,7 @@ void *__wrap__realloc_r(struct _reent *r, void *p, size_t size)
 	return ptr;
 }
 
-void __wrap__free_r(struct _reent *r, void *p)
+void modm_used __wrap__free_r(struct _reent *r, void *p)
 {
 	// do nothing if NULL pointer
 	if (!p) return;

src/modm/platform/spi/stm32/spi_master_dma.hpp.in

@@ -85,9 +85,9 @@ private:
 	static void
 	handleDmaTransmitComplete();
 
-	static inline bool dmaError { false };
-	static inline bool dmaTransmitComplete { false };
-	static inline bool dmaReceiveComplete { false };
+	static inline volatile bool dmaError { false };
+	static inline volatile bool dmaTransmitComplete { false };
+	static inline volatile bool dmaReceiveComplete { false };
 
 	// needed for transfers where no RX or TX buffers are given
 	static inline uint8_t dmaDummy { 0 };

src/modm/platform/uart/nrf/uart_buffer.hpp

@@ -34,7 +34,7 @@ class BufferedUart<Hal, UartTxBuffer<SIZE>, Buffers...>: public BufferedUart<Hal
 	static_assert(not Parent::TxBufferSize, "BufferedUart accepts at most one TxBuffer type");
 	static inline UartTxBuffer<SIZE> txBuffer;
 	static inline uint8_t txBurstBuffer[SIZE];
-	static inline std::size_t txBurstLength = 0;
+	static inline volatile std::size_t txBurstLength = 0;
 
 	static inline void
 	serviceTxBurst()

src/modm/platform/uart/nrfe/uart_buffer.hpp

@@ -34,7 +34,7 @@ class BufferedUart<Hal, UartTxBuffer<SIZE>, Buffers...>: public BufferedUart<Hal
 	static_assert(not Parent::TxBufferSize, "BufferedUart accepts at most one TxBuffer type");
 	static inline UartTxBuffer<SIZE> txBuffer;
 	static inline uint8_t dmaTxBuffer[SIZE];
-	static inline std::size_t dmaTxLength = 0;
+	static inline volatile std::size_t dmaTxLength = 0;
 
 	static inline void
 	serviceDmaTx()

src/modm/processing/fiber/context_arm_m.cpp.in

@@ -253,14 +253,14 @@ modm_context_jump(modm_context_t*, modm_context_t*)
 %#
 		MODM_POP_CONTEXT()
 
-	"1:  b modm_context_end	\n\t"
+	"1:  bl modm_context_end	\n\t"
 %% if not with_psplim
 		:: "i" (StackWatermark)
 %% endif
 	);
 }
 
-void modm_naked
+void modm_naked modm_used
 modm_context_end(uintptr_t)
 {
 	asm volatile

src/modm/processing/fiber/context_avr.cpp

@@ -162,7 +162,7 @@ modm_context_start(modm_context_t *to)
 	return modm_context_jump_entry((modm_context_t*)&main_context_sp, to);
 }
 
-void
+void modm_used
 modm_context_end(uintptr_t retval)
 {
 	modm_context_jump_return(retval, (modm_context_t*)&main_context_sp);

tools/build_script_generator/common.py

@@ -290,8 +290,12 @@ def common_compiler_flags(compiler, target):
         flags["cflags"].append("-g3")
         flags["cxxflags"].append("-g3")
     # flags for the linker
+    flags["linkflags"] = [
+        "-ffunction-sections",
+        "-fdata-sections",
+    ]
     if target.identifier["family"] != "darwin":
-        flags["linkflags"] = [
+        flags["linkflags"] += [
             "-Wl,--fatal-warnings",
             "-Wl,--gc-sections",
             "-Wl,--relax",

tools/build_script_generator/make/resources/compiler.mk.in

@@ -32,16 +32,15 @@ ASM := $(CC)
 
 ifeq ($(C_SUFFIX),)
 	AS := $(C_PREFIX)as
-	AR := $(C_PREFIX)ar
-	NM := $(C_PREFIX)nm
-	RANLIB := $(C_PREFIX)ranlib
 else
 	AS := $(CXX)
-	AR := $(C_PREFIX)gcc-ar$(C_SUFFIX)
-	NM := $(C_PREFIX)gcc-nm$(C_SUFFIX)
-	RANLIB := $(C_PREFIX)gcc-ranlib$(C_SUFFIX)
 endif
 
+_probe = $(if $(shell command -v $(1) 2>/dev/null),$(1),$(2))
+AR := $(call _probe,$(C_PREFIX)gcc-ar$(C_SUFFIX),$(C_PREFIX)ar)
+NM := $(call _probe,$(C_PREFIX)gcc-nm$(C_SUFFIX),$(C_PREFIX)nm)
+RANLIB := $(call _probe,$(C_PREFIX)gcc-ranlib$(C_SUFFIX),$(C_PREFIX)ranlib)
+
 OBJCOPY := $(C_PREFIX)objcopy
 OBJDUMP := $(C_PREFIX)objdump
 SIZE := $(C_PREFIX)size

tools/build_script_generator/module.lb

@@ -74,6 +74,9 @@ def prepare(module, options):
     module.add_option(
         BooleanOption(name="info.build", default=False,
                       description=descr_info_build))
+    module.add_option(
+        BooleanOption(name="lto", default=False,
+                      description=descr_lto))
 
     if platform in ["avr"]:
         module.add_option(
@@ -186,6 +189,10 @@ def build(env):
     for flag, values in common_compiler_flags("gcc", env[":target"]).items():
         env.collect(flag, *values)
 
+    if env["lto"]:
+        env.collect(":build:ccflags", "-flto=auto")
+        env.collect(":build:linkflags", "-flto=auto")
+
     # Copy python tools folder
     platform = env[":target"].identifier["platform"]
     is_cortex_m = env[":target"].has_driver("core:cortex-m*")
@@ -301,6 +308,20 @@ descr_info_git = """# Generate git repository state information
 
 descr_info_build = """# Generate build state information"""
 
+descr_lto = """# Enable Link-Time Optimization
+
+Compiles with `-flto=auto`, which defers code generation to link time so that
+the optimizer can inline and eliminate code across translation units. This
+typically shrinks the binary noticeably at the cost of a slower link step.
+
+!!! warning "LTO cannot see references from the linkerscript or inline assembly"
+    The optimizer decides what is reachable *before* the linkerscript is
+    applied, so `KEEP()` does not protect a symbol from being deleted. Anything
+    that is only referenced by the hardware (like the vector table), by the
+    linkerscript, or from an inline assembly string must be marked `modm_used`.
+    modm does this for its own symbols.
+"""
+
 descr_openocd_cfg = """# Path to a custom OpenOCD configuration file
 
 If you have a custom configuration file for your target, it will get included

tools/build_script_generator/scons/site_tools/gcc_retarget.py

@@ -50,15 +50,12 @@ def generate(env, **kw):
     env['CXX'] = prefix + 'g++' + suffix
     env['AR'] = prefix + 'ar'
     env['RANLIB'] = prefix + 'ranlib'
-    if suffix == '':
-        env['AS'] = prefix + 'as'
-        env['NM'] = prefix + 'nm'
-    else:
-        env['AS'] = prefix + 'gcc' + suffix
-        env['NM'] = prefix + 'gcc-nm' + suffix
-        if sys.platform != "darwin":
-            env['AR'] = prefix + 'gcc-ar' + suffix
-            env['RANLIB'] = prefix + 'gcc-ranlib' + suffix
+    env['AS'] = prefix + 'as' if suffix == '' else prefix + 'gcc' + suffix
+
+    env['NM'] = prefix + 'nm'
+    for var, wrapper in [('AR', 'gcc-ar'), ('RANLIB', 'gcc-ranlib'), ('NM', 'gcc-nm')]:
+        if shutil.which(prefix + wrapper + suffix) is not None:
+            env[var] = prefix + wrapper + suffix
 
     env['OBJCOPY'] = prefix + 'objcopy'
     env['OBJDUMP'] = prefix + 'objdump'

tools/modm_tools/size.py

@@ -76,6 +76,15 @@ def format(source, device_memories):
         else:
             memories["rom"].append(memory)
 
+    for name, mems in list(memories.items()):
+        merged = []
+        for memory in sorted(mems, key=lambda m: m["start"]):
+            if merged and (merged[-1]["start"] + merged[-1]["size"] == memory["start"]):
+                merged[-1]["size"] += memory["size"]
+            else:
+                merged.append(dict(memory))
+        memories[name] = merged
+
     memory_sections = []
     with open(source, "rb") as src:
         elffile = ELFFile(src)